Image Restoration Software. - Ciencias de la Computación e
Contents
1. Image Restoration Software Blur and Noise Application 1 The Load Original Image button allows us to load an image from within the system in order to work with it 2 This space will display the image we loaded before 3 The Blur panel allows user to apply a blur mask to the loaded image 4 The Noise panel allows user to apply noise to the previously blurred image 5 This space will show the observed image first the blurred image and later if this is the case the blurred and noisy image 6 The Save Degraded Image button allows us to save the observed image in the system 7 The Exit button closes the application 2 2 Load an Image First thing we need to start working in order to obtain an observed image is to load an image so we click on the Load an Image button as it is shown in the picture Image Restoration Software Blur and Noise Application EA Blur ndioiso Jm fe Ed Save Degraded image 2 Load Blur 2 Additive Gaussian O Paissan C Generate Blur C ubpicative Gaussian Then we will need to select and open an image using the browser that pops up We are able now to search the system looking for an image to degrade When the image is found the next step is to click on the button Open in the browser as it is shown in the picture FI AlurindHolze 3 Select the Picture 2E Lock int ES proyecto e a LJ cameraman bmp iL henna EMP Elena2. Restored Image The information that we see is the noise image and blur estimates and the restored image This information is refreshed every the set number of iterations that we specified on the options on the previous window 3 5 3 Show Log Panel option selected If this option is selected more information will be shown than with the previous option and the progress bar The window will have this aspect Image Restoration Software Blind Deconvolution Application Blind deconvolution Size 512 x512 Log Panel ES IICN 2018811 8904 trace CcovF 0 Total 9 8438e 010 Ch 2 0 00025805 trace CcovH 0 Total 9 8438e 010 Total 23 5855 Noise variance var_noise 3 5965 Image prior variance var im 27 7012 Blur prior variance var bl 9 8438e 010 Ifnew fold 192289 8817 fold 34352312 4854 Restoring new foldi fold 0 0055976 lIHnew Hold 23 3623 Hold 254 2418 3 5 4 Create a Log file option selected If this option was selected a log file will be created The log file is a txt file with information A typical log file is shown below E log_restoration txt WordPad SEE Archivo Edici n Ver Insertar Formato Ayuda Dc E 664 A 2B X Initial parameters value Noise variance var_noise 7 364054 Image prior variance var_im 7 667433 Blur prior variance var bl 0 000000 g H 2 1930442 468807 Jee e 2
3. 3 1 Main Window Here we can see what the program looks like when it is started and a brief explanation is given of the main buttons In this step we must load an Observed Image an Initial Image and the Initial Blur for the Blind Deconvolution process 6 4 Blind deconvolution Load Initial Blur Size Load Observed Image 5 Load Initial Image A 24 Image Restoration Software Blind Deconvolution Application 1 The Load Observed Image button allows us to load an image from within the system which will be the degraded Observed Image 2 The Load Initial Image button allows us to load an image from within the system which will be the Initial Image for the restoration process 3 The Load Initial Blur button allows us to load an image from within the system which will be the Initial Blur for the restoration process 4 The Exit button closes the application 5 The Next button takes us to the next window of the application in which we can set up the parameters for the Blind Deconvolution process 6 The Reset button takes us to the main window of the application 3 2 Load an Image First we click on the Load Observed Image button in order to load the image we want to restore Image Restoration Software Blind Deconvolution Application Blind deconvolution Size Load Initial Blur Load Observed Image Load Initial Image Then
4. Full size the PSF support will be the same size of the image during the Blind Deconvolution process Components of the PSF could be positive or negative values Set height and width with this option you have to set the size of the PSF support Advanced Option PSF support Set heigth x width gt This option will set to O all the values of the PSF that are outside the kernel size The default value is 20 x 20 Components of the PSF could be positive or negative values but normalization is applied to the PSF in order to keep its components positives and negatives making 1 An example is given here to illustrate the process If we have this PSF and we select a 4x4 PSF support The program will set to O all the values out of the grey squares and then normalize all the PSF in order to keep the values making 1 The program will do this each iteration Automatic with this option the program sets automatically the PSF It sets to 0 all the values from the border of the PSF until it founds a negative value so no negative value will remain in the support Here again it is necessary to normalize the PSF in order to keep the values making 1 Example 33 Image Restoration Software Blind Deconvolution Application We have a PSF like this Positive values Negative values The program will set to O all the values since it founds a negative value starting from the center to the borders The result
5. Generete Blur Save Blur O Multiplicative Gaussimn A saving window will pop up so we are able to specify the name and the extension of the image The extension mat will be given as a default but we are able to specify any other file format by specifying it in the window at the end of the file name u p omm ma mm j uus ndMoise Save as Save in L3 proyecto c er E3 de asussienglu 25 6 mat wu ape ap Een A re Se Sy O Leod Bir Cut 5 Poison 5 Generate Blur y Multiplicative Gaussian ii 22 Image Restoration Software Blur and Noise Application 2 6 Closing the application To close the application and all its windows we just have to press the Exit button as It is shown in the picture Load Blur Additive Gaussian 9 Generebe Bie Er C2 Mutiplicative Gaussier Image Restoration Software Blind Deconvolution Application 3 Blind Deconvolution Application This application implements the Blind image restoration method proposed in 1 To start the module we must run the Blind_deconvolution m file in the Matlab console Another option is to run the Guide application in Matlab and run the Blind deconvolution fig file with it Do not forget to change the Matlab Current Directory in the Matlab console to the one in which the Blind_deconvolution files are stored in order to be able to run it
6. Poisson noise Additive Gaussian noise or Multiplicative Gaussian noise 17 Image Restoration Software Blur and Noise Application e mm ers Sa nn Land Origina Image Bug z Nii 2 Load Blur 5 Additive Gaussian O Poisson 3 Generate Elur I C Multiplicative Gaussian 2 4 1 Poisson noise This type of noise does not require any kind of parameters to be provided so the interface directly will ask us for confirmation by clicking on Apply button as it is shown in the picture 18 Image Restoration Software Blur and Noise Application Losd Original Image Poisson Noise L Porson Moise doesnt require specific parameters After pressing Apply the right image window will be updated with the desired noise added to the blurred image 2 4 2 Additive Gaussian noise In order to introduce this kind of noise we will be asked to provide the noise variance or the signal noise ratio If we press the Enter key in our keyboard after introducing any of the parameters the other will be automatically actualized so we Will see if the noise is the one desired before actually applying it 19 Image Restoration Software Blur and Noise Application Load Original Image Save Degraded Image Addie Gausson hoise Deine Specific Parameters 7 3 zx 2 4 3 Multip
7. Restoration Software ok LS Table of Contents 4 3 2 Calculating the ee Ee EE EE EE ER EE ER ER ER ee ER ee Re ee ee ee 47 ISNR Example of Use EO a 48 Authors of INIS RR 52 Document z BIDIHOGgFaDHV aan DE Ee diae aS Rui 53 Image Restoration Software Introduction 1 Introduction As part of the Computer Engineering Final Project supervised by Prof Rafael Molina in collaboration with Dr Javier Mateos members of the Department of Computer Science and Artificial Intelligence CCIA at the University of Granada Spain we have developed a software package that is able to perform Blind Deconvolution using a variational approach to parameter image and blur estimation 1 This software is divided into three modules the first one degrades a given original image applying blur and noise to it The second module actually performs the Blind Deconvolution The third gives useful information about the deblurring process such as the improvement in signal to noise ratio ISNR as an objective measure of the process quality 1 1 Purpose of this Document The aim of this document is to easily explain how each of the modules works so it would be easy for anyone to use them Each and every one of the features the programs allow us to use will be explained here step by step using many pictures and diagrams First of all the Noise and Blur application will be explained with a series of snapshots of the program itself
8. To apply this blur we will be asked to introduce two parameters that define the blur Size and Sigma This creates a Size x Size Gaussian kernel filter with variance value of Sigma EA BlurandNoise E Land Original Image Save Degraded Image Gasan BA Moise Dele Spacitic Parameters 8 Addie Gaussian Size C Poisson O mutipicative Gaussian Sigma i Apply Rene Current Norge Fai 13 Image Restoration Software Blur and Noise Application 2 3 2 2 Uniform Blur This blur requires a parameter that is the size of the blur kernel and so the interface will ask us for it El BturandNoise Uniform Blur Moke Defire specific Parameters 2 Adctive Gaussian Size y Muttiplicative Gaussian 2 3 2 3 Customized Blur This option gives us the possibility to configure a blur kernel according to our own preferences giving us a 7x7 matrix to specify it The kernel must be centered in the middle of the square 4 row 4 column 14 Image Restoration Software Blur and Noise Application lBiur ndNoise A i Active Gaussian O Mutlipiealive Gaussian 2 3 2 4 Motion Blur To apply this kind of blur we will be asked to introduce two parameters that define the blur Size and Theta The motion blur approximates the linear motion
9. and all the menus and functionalities will be shown oecondly the main program will be explained in a similar fashion covering the full functionality of the Blind Deconvolution Algorithm developed The third part stands for a brief explanation of the last module so it can make clear its purpose and functionality and instructions for use will be given as well The last part of this document contents an example of use that may assist a first time user of the application Image Restoration Software Blur and Noise Application 2 Blur and Noise Application This application performs the functionality to degrade an image with blur and noise determined by the user To start the module we must run the BlurAndNoise m file in the Matlab console Another option is to run the Guide application in Matlab and run the BlurAndNoise fig file with it Do not forget to change the Matlab Current Directory in the Matlab console to the one in which the BlurAndNoise files are stored in order to be able to run it The following sections show how to use the program and describe the functionalities implemented on it 2 1 Main Window Here we can see what the program Blur and Noise looks like when it is started and a brief explication is given of the main buttons and menus El Biur ndNoise Piel x Do gg 9 gd dral Image Q 5 Load Blur 8 Additive Gaussian Generate Blur i idulbipicalive Gaussian
10. of a camera by Size pixels with an angle of Theta degrees in a counter clockwise direction It is so that the user must introduce the Size in pixels and the angle in degrees to obtain correct results 15 Image Restoration Software Blur and Noise Application Motion Bolsa Dating Specific Parameters Blur 25 Additive Gaussian Size 3 Poisson O Multtigkealtive Gaussisn Theia P 2 3 3 Situation after blur application After any of the Blur types have been applied we will show the main Blur panel view again In the next picture we give details about the new situation 16 Image Restoration Software Es Blur and Noise Application Lomi Origins Image CO Load Blur AcidEive Gaussian ri 5 Generate Blur CO Muliplicative Gaussian 1 Here we have the blurred image 2 This Show Blur Image button allows us to see the blur matrix as an image 3 The Remove Current Blur button deletes blur from the image and lets us start from the beginning 4 The Next button in the Noise panel is now active so we can apply noise to the blurred image 5 With the Save Blur button we have the possibility to save the applied blur into a mat file 2 4 Applying noise to an image The second transformation we can do to the image is noise application We now show how to add noise to the blurred image The interface gives us different possibilities
11. uie 3 tif Ellena_unta_40 6f File name y Files of lupe Blur Moise i 5 Load Blur 5 Additive Gaussian 15 Poisson C Mullinteslive Gaussian O Generate Blur Image Restoration Software Blur and Noise Application When selected the image will be shown in the left square of the interface looking like this E Blur ndHoise e Addie Gaussian C Generate Blur C Miuttiplicatve Gaussian Note that the button Next is now available in the Blur panel so we are now able to start working on the image 2 3 Blurring the selected Image We are now going to use the Blur panel to blur the loaded image The panel gives us two options 2 3 1 Load Blur This option allows us to choose a blur mask stored in a mat file to apply it to the loaded image It must be taken into account that the blur kernel must be stored in the first variable stored in the mat file If the file in question has been saved with the Blur and Noise application there will be no problem loading it We have to click on the Next button while the Load Blur option is selected as it is shown in the next picture The center of the blur is the center of the mask Image Restoration Software Blur and Noise Application 9 Addilive Gaussian O Mutiplicatve Gaussian Next thing we have to do is select the desired blur mask using the browser that comes up and click on Next button on the Blur p
12. 009964 000000 IED 0 001205 Iteration g HE 2 933160 trace FcovH trace Hcovfj O trace covHcovF O Total 3 55972 C 2 1 84917e 006 trace CcovF Total 7 05404 Ch 2 0 000362197 trace CcovH Total 1 38168e 009 Noise variance var_noise 3 55972 Image prior variance var im 7 05404 Blur prior variance var bl 1 38168e 009 fnew fold 657504 fold 3 42709e 007 f new fold fold 0 0200696 Hnew Hold 36 7279 Hold 72 2163 Hnew Hold Hold 0 508582 z Para obtener Ayuda presione F1 40 Image Restoration Software Blind Deconvolution Application 3 6 Results When the Blind Deconvolution process finishes we can see the results This final window has the following aspect and functionality Blind deconvolution TI Restored Image Obtained Blur Noise Variance 3 92543 Image Variance sd Blur Variance 020998010 O 1 Observed image and its size in pixels 2 Results panel We can see the restored image the estimated blur and the noise image and blur estimates 3 Save results button It allows the user to save the image the blur and the image and blur covariance 4 The Reset button takes us to the starting window of the application 41 Image Restoration Software Blind Deconvolution Application 5 We can leave the application by clicking on the Exit button 3 7 Sav
13. 2007 Cora Beatriz P rez Ariza Jos Manuel Llamas S nchez IMAGE RESTORATION SOFTWARE Blind Image Deconvolution User Manual Version 1 0 Image Restoration Software ok ES Table of Contents 1 Introduction rr rs 4 1 1 Purpose of this esse EE ER ER ER Ee cnn cnn nn 4 document Blur and Noise ccoo 5 Application 2 1 Main Window EE Ge ee ee ee ee 5 2 2 Load an Image see Ese EE RE RE ER ER EE ER ee EE Re RE Re ee Ee ee ee Re ee 6 2 3 Blurring the selected EE EER EE EE ER EE RE RR ER EE RR ER ee Re 8 Image 231 Load Blur ee ee ee ee ee ee 8 2 3 2 Generate Blur ee 10 2 3 2 4 Gaussian ccoo 11 Blur 2 3 2 2 Uniform ic e 12 Blur 2 3 2 3 CUSTOM csc rrr 12 Blur 2 3 2 4 Motion csse RR Reti 13 Blur 2 3 3 Situation after Lsssseueeeeee m 14 Blur application 2 4 Applying noise to an sees 15 Image 2 4 1 Poisson Noise cssscenRIRRRIIIH IH 15 2 12 Additive armenio VERRE RITE MEE 16 Gaussian Noise 2 4 3 Multiplicative o 17 Gaussian noise 2 4 4 Situation after LLuuuuuttuuueeeeeenm e 18 noise application 2 5 Saving Results eeisusixiiobni dia uc DEERE EER Ge 18 2 6 Closing i m O ed 20 Application Blind Deconvolution ssssen men
14. Original Blur Resulted Blur Initial blur If we select a file and click on the Open button the blur and its size are shown Study results 47 Image Restoration Software Study Results Application e The default region that you can see is of the size of the smaller blur matrix and an additional frame of 3 pixels is added around the blur in order to display it better You can modify this region in the show region panel e The original blur refers to the real blur of the degraded image e The initial blur refers to the blur that we used as initial in the Blind Deconvolution process e The Estimated blur is the blur that we obtain as result in the Blind Deconvolution process 4 2 2 Making a 3D comparison We can compare the blurs in a 3D graph by clicking on the Refresh Comparison button It will show only the selected blurs on the Select blur panel 48 Image Restoration Software Study Results Application Study_results Size 33 x33 Size 33 x33 Size Load Original Blur Load Resulted Blur Load Initial Blur Blur LOTTIHMAISSUII Show region Heigth r Select blur to compare Original Original Blur d EN Resulted ai 2 Resulted Blur gt Initial 3 al 2t C Initial blur E oe 9 y l ll Refresh Comparison As we can clearly see in the picture each three blur functions are superposed in a three dimensional plot to allow us to co
15. Then we add additive Gaussian noise with variance 4 and save the degraded image and the blur Now we start the Blind Deconvolution application and load the degraded image we have just created Blind deconvolution Size 512x512 Load Observed Image Load Initial Image We can keep the default initial blur and image given by the program Now we click on Next button and proceed to set the parameters for example the number of iterations equal to 3 and for the rest of parameters we use the default options In the advanced options we set the PSF support to automatic Blind deconvolution BD Parameters Inverse of the hyperprior of the observation precision parameter 1 Beta Confidence on 1 Beta 9 pen Inverse of the hyperprior of the image precision parameter 1 Alpha Ki Confidence on 1 Alpha T Inverse of the hyperprior of the blur precision parameter 1 Alpha E 512 x512 Confidence on 1 Alpha B Max number of iterations C Show blur noise and image variance Options Stopping Criterion and restored image each C Show log panel Prior Model ICAR z C Create a log file Both Determin Distribution of the image and the blur Advanced Option PSF support Automatic 55 Image Restoration Software Example of Use Now we press Apply and start the Blind Deconvolution process At the end of the restoration we see the following sc
16. abacan R Molina A K Katsaggelos Total Variation Blind Deconvolution Using a variational approach to parameter image and blur estimation P Campisi K Egiazarian Blind image deconvolution Theory and Applications CRC Press 2007 R Molina Introducci n al Procesamiento y An lisis de Im genes Digitales curso impartido en Introducci n a la Rob tica hasta 1998 Dpto de Ciencias de la Computaci n e A Universidad de Granada T Bishop D Barbacan B Amizic T Chan R Molina and A K Katsaggelos Blind Image Deconvolution problem formulation and existing approaches in Blind image deconvolution Theory and Applications P Campisi and K Egiazarian ed CRC 2007
17. anel 10 Image Restoration Software Blur and Noise Application GaussionBiur_25_6 mat Blur Noise 2 Load Blur C Generate Blur O Muliplicative Gaussian 5 Addie Gaussian After this the program will give us the option to apply the selected mask or to go back to the main window and start again the blurring process Ed BlurAndNoise 2 Additive Gaussian Mukipicalivo Gaussian Push Appl to continue or Back to cancel 11 Image Restoration Software Blur and Noise Application If we choose to Apply the blurred image will be shown in the right white square showing that blur has been applied successfully 2 3 2 Generate Blur In case we want to generate a blur mask instead of loading it from a file we have to select the Generate Blur option as it is shown in the picture Bliswrinanoise Woe AS Land Original image Hosa 83 Addie Gaussian C3 Multiplicative Gaussian The interface will show the four possibilities we have for blur which are Gaussian Uniform Customized and Motion and user should select one and then click on Next button 12 Image Restoration Software Blur and Noise Application Jo Ed BiurAndNotse Blur Mose Choose Bir Type m Additive Gaussian 2 Gaussian C Custom O Poisson hiutiplicetree Gaussi C uniform C motion OE GE ue 2 3 2 1 Gaussian Blur
18. ation Software Study Results Application Here we can see what the image comparison window looks like and a brief explanation of the main buttons Load Original Image Load Blurred Image 3 Load Restored Image Back Exit 1 By clicking on the Load Original Image we are able to select a picture that represents the original image zem ISNR Cal ulat F 2 By clicking on the Load Blurred Image we are able to select a picture that represents the observed image 3 By clicking on the Load Restored Image we are able to select a picture that represents the restored image 4 The Calculate ISNR button allows you to calculate the ISNR of the restored image 5 The back button takes you to the main window of the application 6 The exit button closes the application 4 3 1 Loading the images Image Restoration Software Study Results Application lf we click on any of the Load buttons a window to select the image file will be opened Study results Select the Picture PIE Buscar en E Blind deconvloution y4 e Ck Eg gl GaussianBlur 25 3 A GaussianBlur 25 6 ig lena_GaussianBlur_25_5 gl mri_GaussianBlur_25_5 Tipo Mat files mat Cancelar En Load Original Image Load Blurred LosiBuredmege Load Restored Load Restored image QN ISNR Calculate ISNR If we select a file and click on th
19. e Open button the selected image is shown Study_results Load Blurred Image Load Restored Image Calculate ISNR 4 3 2 Calculating the ISNR Image Restoration Software Study Results Application Once the three images are loaded it is possible to calculate the ISNR using the following calculation ISNA m 10 log s me Where stands for the original image 7 is the average of the image and 8 amp 5 Stands for the root mean square error The next step is to click on the calculate ISNR to visualize the numeric value The state of the window after this is shown in the next picture Study results Load Original Image Load Blurred Image Load Restored Image Il 1 78264 gt ISNR Calculate ISNR iia 5 Example of Use 53 Image Restoration Software Example of Use To represent the functionality of the software we present here an example of use First of all we degrade an image using the BlurAndNoise application litura ndioise Load Original Image flr JJTdj Z2 gt gt KK Noise O Lead Blur BlurAndNoise EM EM Image Load Blur Show Blur Image Poisson Next Remove Current Noise Remove Current Blur HEH 2 Addilive Gaussian 7 Mubpicative Gaussian otov otov Image let 6 Additive Gaussian Multiplicative Gaussian 54 Image Restoration Software Example of Use
20. ed PSF will be like this Positive values Negative values The program will do this every iteration 3 4 Starting the restoration Once we have set all parameters we click on the Apply button We can select the different options to see information on the iterative process 36 Image Restoration Software Blind Deconvolution Application Blind deconvolution BD Parameters Inverse of the hyperprior of the observation precision parameter 1 Beta 9 Confidence on 1 Beta 0 g Inverse of the hyperprior of the image precision parameter 1 Alpha d Confidence on 1 Alpha l Inverse of the hyperprior of the blur precision parameter 1 Alpha H 512x512 es Confidence on 1 Alpha bl Psf support Max number of iterations Options Stopping Criteria td Show blur noise and image variance Prior Model CAR and restored image each J iterations ter i aa y istributi i Both Determin C Show log panel Distribution of the image and the blur z II C Create a log file If we choose the Create a log file option a window to select the output file pops up Blind deconvolution gt Save log file as Guardar en E Blind deconv V3 gt e ek Ese verse of the hyperprior of the observation precision parameter 1 Beta 0 Confidence on 1 Beta 0 Inverse of the hyperprior of the image precision parameter 1 A
21. ing the results We can save the results by clicking on the Save Results button Blind deconvolution Results Size 512x512 Restored Image Obtained Blur Noise Variance neuen Image Variance bear Blur Variance 4 02755e 010 After this a window to select the name of the file where we are going to save the restored image is opened The default name to save the file is the name of the Observed image plus the value of the parameters and the string rest im The selected name will be the same for the obtained blur blur covariance and image covariance files by adding the strings blur covar blur and covar restored respectively 42 Image Restoration Software Blind Deconvolution Application Blind deconvolution ks Save as Guardar en Gd Blind deconv V3 df DEE Nombre Tipo Restored Image Obtained Blur Noise Variance 3 82543 Image Variance 12 7507 Blur Variance 4 02755e 010 Save Results Here we can see the saved files with their names Blind deconv V3 Archivo Edici n ver Favoritos Herramientas Ayuda OC atr s M J re B squeda Carpetas LE Nombre de Tareas de archivo y carpeta a Crear nueva carpeta Ed Publicar esta carpeta en Web Ed Compartir esta carpeta EK Tama o 1 920 KB 1 990 KB z KB KB 43 Image Restoration Software ES Blind Deconvolution Application 4 Stud
22. itial Blur Load Observed Image Load Initial Image 29 Image Restoration Software Blind Deconvolution Application 3 3 Parameters Window 3 3 1 General Description Here we can see what the window looks like A brief description now is given of the main elements Blind deconvolution 3 BD Parameters Inverse of the hyperprior of the observation precision parameter 1 Beta Confidence on 1 Beta 0 4 Inverse of the hyperprior of the image precision parameter 1 Alpha E i as Confidence on 1 Alpha iml Inverse of the hyperprior of the blur precision parameter 1 Alpha 512x512 Confidence on 1 Alpha bl 0 tions Max number of iterations 50 L Show blur noise and image variance Stopping Criterion 5e 5 and restored image each Iterations Show log panel Prior Mode CAR C Create a log file Distribution of the image and the blur Both Determin Advanced Option 4 5 1 Here we can see the Observed Image and its size 2 The Options Panel has different options that we can choose to show information of the restoration process For further information see chapter 3 3 3 30 Image Restoration Software Blind Deconvolution Application 3 The BD Parameters panel has fields with the different parameters which are used by the Blind Deconvolution algorithm For fur
23. licative Gaussian noise This type of noise requires the parameter that multiplied by the blurred observation corresponds to the variance of the noisy observation so it will be requested through the interface as it is shown below B Blur ndHoise Fa n n bags Multiplicetree Gaussen hloize Deine Spectic Parameters ers sem Biur m el Remove Current Blur 20 Image Restoration Software Blur and Noise Application 2 4 4 Situation after noise application After applying any noise to the blurred image the interface will look like the image below Note that we will not be able to introduce any other kind of noise into the image unless the Remove Current Noise button is clicked on in that case the interface will delete the current noise and we will be able to start from the beginning of the noise application process 2 Lead Blur O Additive Gaussian 9 Generete Blur O Multiplicative Gaussinn p 2 5 Saving Results We are able to save the degraded image at any time in the degradation process only by pressing the Save Degraded Image button as it is shown in the picture Keep in mind that the image being saved is the one showed in the right square of the interface EX Image Restoration Software Blur and Noise Application BY BlurAndMolse de ae ar Pd HERE OE TEN em em a pe UTE Ed dol gor arca SEER IAN rd EE EE 5 Polson 5
24. lpha l Confidence on 1 Alpha RH Inverse of the hyperprior of the blur precision parameter 1 Alpha 5 Confidence on 11 Alpha bl Psf support Max number of iterations Options Stopping Criteria O Show blur noise and image variance i car and restored image each Iterations Prior Model 4 Distribution of the image and the blur Both Determin C Show log panel v Create a log file l We must write the name that we want for the log file and click on the save button Image Restoration Software Blind Deconvolution Application 3 5 During the restoration process Here we will see the result of choosing different options We show the options one by one but we can combine two or more options 3 5 1 No option selected After clicking on the Apply button we will see a window with information If we did not choose any option then the application shows only a progress bar Blind deconvolution Size 512 x512 3 5 2 Show blur noise and image variance and restored image 38 Image Restoration Software Blind Deconvolution Application If this option is selected we will see some information and the progress bar The window will look like the one below Blind deconvolution Supervision Iteration 1 Noise Variance 355872 Image Variance 705404 Size 512 x512 Blur Variance 1 38168e 003
25. lue for this parameter is 0 4 Confidence on the previous parameter Must be a value between 0 and 1 with O meaning no confidence on the given parameter 1 4 phas that is the value of the parameter will be estimated from the data and 1 meaning that it fully 32 Image Restoration Software Blind Deconvolution Application enforces the value of the parameter e no estimation is performed on the parameter 5 Inverse of the hyperprior of the Blur precision parameter 1 Alpha The default value for this parameter is 0 6 Confidence on the previous parameter Must be a value between 0 and 1 with O meaning no confidence on the given parameter 1 4lpha that is the value of the parameter will be estimated from the data and 1 meaning that it fully enforces the value of the parameter i e no estimation is performed on the parameter 7 Maximum number of iterations of the algorithm The process stops if the stopping criterion is met If not it stops when it reaches the maximum number of iterations 8 Stopping criterion The algorithm stops if the result of divide the norm of the difference between the restored image of the actual iteration and the restored image in the previous iteration by the norm of the restored image of the actual iteration is less than this value perm Mathematically Es Stopping criterion 9 Prior model of the image and the blur We can choose between a Simultaneous Auto Regres
26. mpare the blurs in every direction Black color stands for the Original Blur red for the estimated Blur and Green for the Initial Blur used in the Blind Deconvolution process We are able to rotate the 3D graph by moving the mouse pointer while the left button is pressed 4 2 3 Modifying the shown region We can also zoom in or zoom out the blurs using the panel shown in the next picture We can set the size of the region by setting its height and width Then if we click on the Apply button the new region is shown 49 4 Image Restoration Software Study Results Application Study results Size 20 x20 Size 20 x20 Size Load Original Blur Load Resulted Blur Load Initial Blur Blur con Paris Un Show region Heigth Select blur to compare Original 20 Il Original Blur Resulted Resulted Blur Initial O Initial blur T Refresh Comparison 0 ERA eere TN a SS A AROS SN TN US Di Note that the new psf regions will not be displayed in the 3D plot until we click on the Refresh Comparison button Study results Size i Size Load Original Blur Load Resulted Blur Load Initial Blur Blur curriparissuri Show region Heigth Select blur to compare Original Original Blur Resulted 5 Resulted Blur Initial O Initial blur Refresh Comparison 4 3 Image comparison Image Restor
27. n 21 Application 3 1 Main Window Ge ee 21 si Image Restoration Software Table of Contents 3 2 Load an Image nm 22 3 3 Parameters Window sseessseeeeeerrrre 26 3 3 1 General EER NE OO Re OR RE di NE RES 26 Description 3 3 2 Blind ER ER oe OE ES EE ee ee od Ee en EG dE 27 Deconvolution Parameters 393 UDPDU0NS restan ia 29 3 3 4 Advanced eise ee ee ee RR ee Re ee nn nn rra nnne 30 Options 3 4 Starting EE N EE less 32 Restoration 3 5 During ie siene SE err err cre EE ie Te 33 Restoration Process 3 5 1 No ei osse SE OE N DE De de yen 33 Selected 3 5 2 Show Blur Noise le EE orde 34 Variance and Restored Image 3 5 3 Show LOOG AE RE ER EE EK 34 Panel option selected 3 54 Create LOG Fil ses OE Ee RE GE 35 option selected 3 6 Results ra 36 3 7 Saving the Results ooo esse ese EE cece EER Re EE s 37 4 Study A ius Ee EE Ie ee ee re 39 Application 4 1 Main Window sesse ee EE Ee ER RR Ee ee EE RR RR ee ee ee RR ee ee 39 4 2 Blur Comparison see EER EE EE RE ER ee eee ee ee ER ee ee ee ee ee Ge 40 4 2 1 Loading Blurs ies DREEF de EE Gee EG EE GR ER GE GE 41 4 2 2 Making sy 1 O TE EE EE N 42 comparison 4 2 3 Modifying the ooooocccccocccccccccncco nn ee ee s 43 shown region 4 3 1mmage Comparison oo eee eee eee ences zs 45 4 3 1 Loading i esac oO sean 46 images Image
28. nal Blur Load Resulted Blur Load Initial Blur r Select blur to compar Initial blur e e 1 There are three areas where you can load different blurs You will see the size of each blur in pixels in the white rectangles 45 Image Restoration Software Study Results Application 2 3 4 5 6 7 The show region panel allows you to make zoom in or zoom out in the blurs This panel allows you to select the blur or blurs that you want to plot see 4 Here is where the application shows the 3D graph of the selected blurs when you click on the refresh comparison button The back button takes you to the main window of the application The exit button closes the application The Refresh Comparison button shows a 3D graph with the selected blurs in the graphics region see 3 and 4 4 2 1 Loading the blurs If we click on any of the Load buttons a window to select the blur file will be opened 46 Image Restoration Software Study Results Application Study results Select the Picture Buscar en 5 Blind deconvloution v4 Y 153 GaussianBlur_25_3 al GaussianBlur 25 6 gl lena GaussianBlur 25 5 al mri GaussianBlur 25 5 E e E3 Tipo Mat files mat Cancelar i Size bize Load Original Blur Il Load Resulted Blur cu Load zl Show region Heigth Width ee Select blur to compare
29. reen Blind deconvolution Results 512 x512 Restored Image Obtained Blur Noise Variance iind Image Variance eee Blur Variance 1 101296 010 Finally we analyze the restored image and the estimated psf using the third application Study Results We start by visualizing the blur images Study_results Size 33 x33 Size 33 x33 Size Load Original Blur Load Resulted Blur Load Initial Blur Blur curriparissur Show region Heigth Width 33 x 33 Select blur to compare Original Original Blur Resulted Resulted Blur ER ot initial Initial blur Refresh Comparison Image Restoration Software Example of Use Now and just to finish we can calculate the ISNR by pressing the Calculate ISNR button Load Original Image Load Blurred Image Load Restored Image 1 44517 Calculate ISNR We can see that we have obtained an ISNR of 1 446 oT Image Restoration Software Authors of this Document 6 Authors of this Document Jos Manuel Llamas S nchez Cora Beatriz P rez Ariza 58 29 Image Restoration Software Bibliography 8 Bibliography 1 2 S 4 5 R Molina J Mateos and A K Katsaggelos Blind Deconvolution using a variational approach to parameter image and blur estimation IEEE Trans on Image Processing vol 15 no 12 3715 3727 December 2006 S Derin B
30. sive prior model and a Conditional Auto Regressive prior model 10 Distribution of the image and the blur We can choose between Both deterministic Deterministic distribution of the blur and Random distribution of the image Both Random 3 3 3 Options The options of this panel enable us to see information about the Blind Deconvolution process We can run the Blind Deconvolution method without any log information or select one or more options Image Restoration Software Blind Deconvolution Application Options Ta Show Blur noise and image variance and restored image each sm Show log panel 457 Create a log file Advanced Option 1 With this option the application displays the image blur and noise variance and it also shows you the restored image This information is refreshed every given number of iterations See 2 2 Number of iterations to refresh the information in 1 3 With this option the application creates a panel where all the information that the application generates during the restoration process is displayed 4 With this option the application creates a file where all the information that the application generates during the restoration process is stored 3 3 4 Advanced Options This panel inside the Option panel allows us the possibility of choosing between three different PSF sizes 34 Image Restoration Software Blind Deconvolution Application
31. ther information see chapter 3 3 2 4 The Back button takes us to the previous window of the application 5 The Apply button starts the Blind Deconvolution process with the selected parameters and options 3 3 2 Blind Deconvolution Parameters Here we can see the different parameters and a short description of each of them 31 Image Restoration Software Blind Deconvolution Application EL Parameters Inverse of the hyperprior of the observation precision parameter 1 Beta DO 1 Confidence on 1 Bata DO O Inverse af the hyperprior of the image precision parameter VAlpha 0 3 Confidence an 17 Alpha DO 4 Im Inverse of the hyperpriar of the blur precision parameter 1 Alpha 9 5 EE P Confidence on 1 Alpha y EET 09 Max number af iterations Eum stopping Criterion Prior me Distribution of the image and the ur EE Determin 1 Inverse of the hyperprior of the observation precision parameter 1 B The default value for this parameter is 0 2 Confidence on the previous parameter Must be a value between 0 and 1 with O meaning no confidence on the given parameter 1 B that is the value of the parameter will be estimated from the data and 1 meaning that it fully enforces the value of the parameter i e no estimation is performed on the parameter 3 Inverse of the hyperprior of the Image precision parameter l AlphaL The default va
32. we will need to select and open a picture using the browser that pops up Allowed file formats are mat and common image formats like bmp tif png gif or jpg When the image is selected the next step is to click on the button Open in the browser as it is shown in the picture 26 Image Restoration Software Blind Deconvolution Application Blind deconvolution Select the Picture Buscar en EB Blind deconv V3 fa ex E3 Tipo MAT files 7 mat Cancelar Size Load Initial Blur Load Observed Image Load Initial Image When selected the image will appear in the square over the button By default the Initial Image for the restoration process will be the same image and the Initial Blur will be a Gaussian Blur with deviation Sigma 2 The size in pixels of the Observed Image is shown below the image Ps Image Restoration Software Blind Deconvolution Application Blind deconvolution Size 512 x512 Load Initial Blur Load Observed toadcbserved image Load Initial ECT uN We can load a different Initial Image by clicking on the Load Initial Image button Blind deconvolution Load Initial Blur 28 Image Restoration Software Blind Deconvolution Application We can load a different Initial Blur by clicking on the Load Initial Blur button Blind deconvolution Size 512x512 Load Initial Blur Load Observed Image Size 512x512 Load In
33. y results Application This application has been designed to help users of Blind Deconvolution processes to test results obtained It is divided into two main functionalities that are Blur comparison and Calculation of the ISNR as a measure of the quality of the restoration process To start the module we must run the Study results m file in the Matlab console Another option is to run the Guide application in Matlab and run the Study results fig file with it Do not forget to change the Matlab Current Directory in the Matlab console to the one in which the otudy results files are stored in order to be able to run it 4 1 Main Window This program allows us to examine and compare results obtained with the previous two applications The window that first appears when the program is started displays a menu with three options as it is shown in the picture below 0 Q 44 Image Restoration Software Study Results Application 1 The Blur button takes us to the blur window In that window we can load different blurs and see the differences between them 2 The Image button takes us to the image window In that window we can load three images show them and calculate the ISNR 3 The Exit button closes the application 4 2 Blur Comparison It is shown here the blur window appearance and a brief explanation of the main buttons is also given Study_results 1 ize toed origina har ad Origi
Download Pdf Manuals
Related Search